About the Role

Focus on Mirelo AI's full training stack — profiling GPU behavior, debugging training pipelines, improving throughput, choosing parallelism strategies, and designing the infrastructure that lets the team train generative models efficiently at scale.

Key Responsibilities

Find ideal training strategies (parallelism approaches, precision trade-offs) for various model sizes and compute loads
Profile, debug, and optimize single and multi-GPU operations using tools like Nsight and stack trace viewers
Analyze and improve the whole training pipeline (data storage, loading, distributed training, checkpointing, logging)
Set up scalable systems for experiment tracking, data/model versioning, and experiment insights
Design, deploy and maintain large-scale ML training clusters running SLURM for distributed workload orchestration

Requirements

Hands-on experience implementing the latest techniques in optimizing training and inference workloads
Deep understanding of GPU memory hierarchy and computation capabilities
Experience optimizing for both memory-bound and compute-bound operations
Expertise with efficient attention algorithms and their performance at different scales

Details

Hybrid role based in Berlin or Tübingen, Germany. Mirelo AI recently closed a $41M Seed round co-led by Andreessen Horowitz and Index Ventures.

Training Infrastructure Engineer